The SO(3) manifold of 3D rotations. Uses unit quaternions as the representation. Internally converts to/from scipy Rotation objects for Log/E

Generalized Procrustes Analysis (GPA). Algorithm outline: 1. Choose a reference shape. 2. Apply Procrustes Analysis to superimpose all s

Type of variables in an MFA group. A group must be homogeneous: every column in it is either numerical (PCA-style) or categorical (MCA-style)

Multiple Correspondence Analysis (MCA). MCA extends correspondence analysis to more than two categorical variables. It works by one-hot encod

Abstract base class for Riemannian manifolds.

Principal Component Analysis (PCA). Parameters ---------- rescale_with_mean Whether or not to subtract each column's mean before

Principal Geodesic Analysis (PGA). PGA generalizes PCA to Riemannian manifolds. Data is mapped to the tangent space at the Fréchet mean, PCA

Compare PGA results against geomstats TangentPCA as a reference.

MFA on the FactoMineR ``poison`` dataset with one numerical and three categorical groups. Exercises mixed numeric/categorical MFA (FactoMineR's `

random_state must produce reproducible SVDs regardless of the global numpy random state, for every engine that exposes a random_state knob.

Reproduce the example from the Wikipedia FAMD article. https://en.wikipedia.org/wiki/Factor_analysis_of_mixed_data Reference values computed